The default ABIs among mips64{,el} GCC are inconsistent

[Rongronggg9]

$ cd build/latest
$ grep CT_ARCH_mips_ABI mips64*
mips64-11.2.0.config:CT_ARCH_mips_ABI="64"
mips64-12.1.0.config:CT_ARCH_mips_ABI="64"
mips64-12.2.0.config:CT_ARCH_mips_ABI="64"
mips64-12.3.0.config:CT_ARCH_mips_ABI="64"
mips64-13.1.0.config:CT_ARCH_mips_ABI="64"
mips64-13.2.0.config:CT_ARCH_mips_ABI="64"
mips64-4.9.4.config:CT_ARCH_mips_ABI="n32"
mips64-5.5.0.config:CT_ARCH_mips_ABI="n32"
mips64-9.5.0.config:CT_ARCH_mips_ABI="n32"
mips64el-12.1.0.config:CT_ARCH_mips_ABI="32"
mips64el-12.2.0.config:CT_ARCH_mips_ABI="32"
mips64el-12.3.0.config:CT_ARCH_mips_ABI="32"
mips64el-13.1.0.config:CT_ARCH_mips_ABI="32"
mips64el-13.2.0.config:CT_ARCH_mips_ABI="32"
mips64el-4.9.4.config:CT_ARCH_mips_ABI="n32"
mips64el-5.5.0.config:CT_ARCH_mips_ABI="n32"
mips64el-9.5.0.config:CT_ARCH_mips_ABI="n32"

Is this intended? In theory, all 64-bit MIPS targets should default to the N64 ABI, which should set CT_ARCH_mips_ABI="64".

If not, I am willing to open a PR and fix it.

[Rongronggg9]

Furthermore, configurations of the same GCC version should be consistent except for the endianness. However, only the recently-merged configurations, which are mips64{,el}-{4.9.4,5.5.0,9.5.0}.config, fit the above assumption, while the others are not. The difference between, for example, mips64{,el}-13.2.0.config, is so huge that even floating point (hard/soft) and libc are different.

Should it be a mistake, we may need to make a further fix.

[maringuu]

According to this comment this is neither intended nor unintended.
It simply is this way because nobody cared.

I personally would be happy if the configurations were consistent.

[pinskia]

Note most mips64-linux-gnu gcc cross compilers default to n32 (which is the default one without supplying an extra option to configure) , the only ones I know of which don't were the octeon ones but that was over 5 years ago so things might have changed since then.